Leaf Only Sam

[Paper] [Data]

This is the code from the paper Leaf Only SAM: A Segment Anything Pipeline for Zero-Shot Automated Leaf Segmentation

The is a series of post processing steps to be applied to the output of Segment Anything to identify only leaf masks. The method has been developed and tested on a novel dataset of potato leaf images which is located here

Installation

In order to use it you need to have segment anything working on your system and opencv.

The code requires python>=3.8, as well as pytorch>=1.7 and torchvision>=0.8. The easiest way to set up is to create a conda environment by running the command.

conda create --name leafonlysam numpy pytorch torchvision torchaudio pytorch-cuda=12.4 opencv matplotlib pandas ipykernel -c pytorch -c nvidia

You will then need to add in segment anything by running

pip install git+https://github.com/facebookresearch/segment-anything.git

Citing Leaf Only SAM

If you use Leaf Only SAM in your research, please cite the following BibTeX entry.

@misc{williams2023leaf,
      title={Leaf Only SAM: A Segment Anything Pipeline for Zero-Shot Automated Leaf Segmentation}, 
      author={Dominic Williams and Fraser MacFarlane and Avril Britten},
      year={2023},
      eprint={2305.09418},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}